We introduce an intraday co-jump test measuring the alignment of intraday candlesticks by utilising thresholding and cosine similarity. Based on the distributional properties of cosine similarity, we derive a test that operates on a rolling window basis, making it possible to detect co-jumps on an interval basis in a near-real-time manner (e.g. 5-minute, 1-minute and 30-second sampling). We conduct an extensive scenario-based Monte Carlo simulation study to evaluate the proposed methodology under market microstructure noise and asynchronous trading. We replicate a set of key existing co-jump tests to measure the comparative performance of our test. The results demonstrate that our approach consistently outperforms existing methodologies in detecting moderate and small co-jumps, showing a higher degree of robustness to noise and asynchronicity, both daily and intraday. Our empirical analysis indicates that co-jumps are likely more frequent than detected by existing methodologies, both for scheduled announcements and news events.